Research map / Probabilistic and scientific modeling / Variational and generative sampling
BIVA: A Very Deep Hierarchy of Latent Variables for Generative Modeling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add skip connections (deterministic top-down path) (ablation) | LVAE · CIFAR-10 | bits/dim: 3.60 -> 3.41 | Helped |
| add bidirectional inference network (ablation) | LVAE+ · CIFAR-10 | bits/dim: 3.41 -> 3.12 | Helped |
| increase stochastic depth L=15 -> L=29 (ablation) | LVAE+ · CIFAR-10 | bits/dim: 3.41 -> 3.45 | Hurt |
| increase layers L=10 -> L=15 (ablation) | BIVA · CIFAR-10 | bits/dim: 3.17 -> 3.12 | Helped |
| finetune with free bits lambda=0 (ablation) | BIVA · Static binarized MNIST | NLL: 81.20 -> 80.47 | Helped |
| BIVA instead of best flow model | non-autoregressive generative models · CIFAR-10 | bits/dim: 3.08 -> 3.08 | No clear effect |
| BIVA instead of autoregressive PixelCNN++ | generative models · CIFAR-10 | bits/dim: 2.92 -> 3.08 | Hurt |
| use high-level latent ELBO L>k instead of standard ELBO for OOD scoring | BIVA · CIFAR-10 vs SVHN | bits/dim: SVHN 2.28 vs CIFAR 3.12 (k=0) -> 121.04 vs 79.36 | Helped |
Closest papers by meaning
- Trading Information between Latents in Hierarchical Variational Autoencoders (ICLR 2023 · similarity 0.88)
- Variational Autoencoders with Jointly Optimized Latent Dependency Structure (ICLR 2019 · similarity 0.88)
- DVAE++: Discrete Variational Autoencoders with Overlapping Transformations (ICML 2018 · similarity 0.88)
- Deep Attentive Variational Inference (ICLR 2022 · similarity 0.88)
- NVAE: A Deep Hierarchical Variational Autoencoder (NeurIPS 2020 · similarity 0.88)
- EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.